Commit Graph

876 Commits

Author SHA1 Message Date
ameerj
67159e3be7 dead_code_elimination_pass: Remove unreachable Phi arguments 2022-03-23 17:57:22 -04:00
Billy Laws
dd862dff66 Include <bit> header when std::count{r,l}_zero is used
Needed for compilation with older libc++ releases
2022-03-22 21:11:24 +00:00
ameerj
f10d40a0a2 shader_recompiler/dead_code_elimination: Add DeadBranchElimination pass
This adds a pass to eliminate if(false) branches within the shader code
2022-03-22 02:39:31 -04:00
Ameer J
75046a3351
Merge pull request #8038 from liamwhite/exit-register-detection
shader_recompiler/EXIT: increment output register on failed enable test
2022-03-21 21:24:07 -04:00
ameerj
866b7c20a8 general: Fix clang/gcc build errors 2022-03-20 02:25:09 -04:00
ameerj
574a2c4b77 shader_recompiler: Reduce unused includes 2022-03-20 02:25:08 -04:00
Liam
536d7ed7b1 Address review comments 2022-03-18 15:55:46 -04:00
Liam
d400b618a7 shader_recompiler/EXIT: skip render targets with no outputs 2022-03-18 09:26:25 -04:00
Liam
6fa17f3372 shader_recompiler/EXIT: increment output register on failed enable test 2022-03-17 22:09:31 -04:00
Liam
3009d0bd7d Address review comments 2022-03-17 14:48:18 -04:00
Liam
e228a40db8 shader_recompiler: Use functions for indirect const buffer accesses 2022-03-17 13:30:21 -04:00
Liam
3ac522ba41 Address review comments 2022-03-17 09:30:41 -04:00
Liam
1415542f73 shader_recompiler: Implement LDC.IS address mode 2022-03-16 11:05:04 -04:00
Liam
52895fab67 shader: add support for const buffer indirect addressing 2022-03-14 19:43:32 -04:00
Fernando S
cd07a43724
Merge pull request #8008 from ameerj/rescale-offsets-array
rescaling_pass: Fix rescaling Color2DArray ImageFetch offsets
2022-03-15 00:08:22 +01:00
Fernando Sahmkow
185fc03c3c Shader decompiler: do constant propgation before texture pass. 2022-03-13 21:49:40 +01:00
Fernando Sahmkow
ec9f0f064e Shader decompiler: Fix storage tracking in deko3d. 2022-03-13 17:41:16 +01:00
ameerj
f87f8d4610 rescaling_pass: Fix rescaling Color2DArray ImageFetch offsets
ImageFetch offsets for 2D array coordinates have a different composite size than the coordinates. The rescaling pass was not taking this into account.

Fixes broken shaders when scaling is enabled in Astral Chain, and likely other titles.
2022-03-12 03:31:56 -05:00
ameerj
e8c50e709e emit_spirv, vk_compute_pass: Resolve VS2022 compiler errors 2022-03-12 02:54:33 -05:00
Markus Wick
c78c8190d5 shader_recompiler/LOP3: Use brute force python results within switch/case.
Thanks to @asLody for optimizing this function. This raised the focus that this function should be optimized more.

The current table assumes that the host GPU is able to invert for free, so only AND,OR,XOR are accumulated in the performance metrik.

Performance results:

Instructions
0: 8
1: 30
2: 114
3: 80
4: 24

Latency
0: 8
1: 30
2: 194
3: 24
2022-03-08 09:44:28 +01:00
ameerj
e394e1ecc4 emit_glsl_atomic: Implement 32x2 fallback atomic ops 2022-01-29 19:56:03 -05:00
ameerj
90a0506d56 lower_int64_to_int32: Add 64-bit atomic fallbacks 2022-01-29 19:56:02 -05:00
ameerj
ad58d7eae7 shaders: Add U64->U32x2 Atomic fallback functions 2022-01-29 19:55:53 -05:00
ameerj
4790ba7839 spirv_atomic: Define U32x2 storage buffers for 64-bit storage atomics
Some drivers do not support 64-bit atomics, and fallback to atomically modifying U32x2 vectors. This change ensures that U32x2 storage vectors are defined in the spir-v shader when 64-bit atomics are used.

Fixes a hang on some devices, notably Intel GPUs, when booting Pokemon Legends Arceus
2022-01-28 19:00:04 -05:00
Morph
8dea7fa129
Merge pull request #7786 from ameerj/vmnmx-sel
video_minimum_maximum: Implement src operand selectors
2022-01-28 18:24:56 -05:00
ameerj
beaf7654bb emit_spirv: Add Xfb execution mode when transform feedback is used
Fixes Transform Feedback on Vulkan AMD drivers.
2022-01-28 16:32:48 -05:00
ameerj
74e6e3623f video_minimum_maximum: Implement src operand selectors
Used by Pokemon Legends: Arceus
2022-01-27 14:55:08 -05:00
Lioncash
b46ec4efea shader_recompiler: Remove unnecessary [[nodiscard]]
Since ConvertLegacyToGeneric has a void return value, there's nothing
that is actually returned by the function.
2022-01-25 12:16:09 -05:00
v1993
a943600019
shader_recompiler: fix potential OOB access
Found by static analysis with PVS-Studio. Original check wasn't actually checking for OOB and would segfault in case of it.
2022-01-17 21:50:51 +03:00
liushuyu
09f4f3f23b
logging/log.h: move enum class formatter to a separate file ...
... to common/logging/formatter.h
2022-01-09 17:35:33 -07:00
liushuyu
42f653ab6f
logging: adapt to changes in fmt 8.1 2022-01-08 01:49:26 -07:00
Narr the Reg
524674aafc glsl: Remove unreachable return 2022-01-04 20:23:39 -06:00
Fernando Sahmkow
f58ee3f15f ShaderDecompiler: Add a debug option to dump the game's shaders. 2022-01-04 02:39:00 +01:00
Fernando S
ae7da0b12d
Merge pull request #7629 from ameerj/nv-driver-fixes
shaders: Add fixes for NVIDIA drivers 495+
2022-01-03 00:39:59 +01:00
ameerj
8c907c620d glsl: Add boolean reference workaround 2021-12-29 19:03:50 -05:00
ameerj
b84d429c2e glsl_context_get_set: Add alternative cbuf type for broken drivers
some drivers have a bug bitwise converting floating point cbuf values to uint variables. This adds a workaround for these drivers to make all cbufs uint and convert to floating point as needed.
2021-12-29 19:03:50 -05:00
ameerj
9f34be5a61 emit_glsl_integer: Use negation work around 2021-12-29 19:03:50 -05:00
ameerj
14ac0c2923 shader: Add integer attribute get optimization pass
Works around an nvidia driver bug, where casting the integer attributes to float and back to an integer always returned 0.
2021-12-29 19:03:49 -05:00
bunnei
279c7bcc1a
Merge pull request #7618 from goldenx86/patch-4
Increase boost requirement to 1.78.0
2021-12-28 16:25:37 -08:00
Matías Locatti
c7235e67ef
Empty spaces 2021-12-28 18:50:51 -03:00
Matías Locatti
840d5520d2
Changes to avoid warnings in SSE4.2 optimized SPIR-V 2021-12-28 17:35:55 -03:00
ameerj
37addf7a94 emit_glasm_context_get_set: Fix GetAttribute return value type.
GetAttribute expects an F32 result type at the IR level, this fixes the return value of attributes which were not returning an F32
2021-12-24 20:45:07 -05:00
ameerj
640fc1418b emit_glsl_floating_point: Fix FPNeg on newer Nvidia drivers 2021-12-24 20:03:54 -05:00
vonchenplus
4908a07c20 Address format clang 2021-12-18 14:27:07 +08:00
vonchenplus
6ebc972c2b Remove spirv handle legacy related code 2021-12-18 14:08:50 +08:00
vonchenplus
94652e122d Remove glsl handle legacy related code 2021-12-18 14:03:40 +08:00
Feng Chen
e49184e606
Merge branch 'yuzu-emu:master' into convert_legacy 2021-12-18 13:57:14 +08:00
Mai M
edbde7a220
Merge pull request #7522 from ameerj/shader-recompiler-filenames
shader_recompiler/backend: Minor organization and refactoring to reduce compile time overhead
2021-12-07 18:27:50 -05:00
ameerj
7105204a7e emit_spirv: Reduce emit_spirv.h include overhead
emit_spirv.h is included in video_core, which was propagating further includes that video_core did not depend on.
2021-12-05 18:11:19 -05:00
ameerj
1e1f7b3234 glasm: Move implemented instructions from not_implemented.cpp 2021-12-05 18:11:19 -05:00
ameerj
f32b2bcd20 shader_recompiler: Adjust emit_context includes 2021-12-05 18:11:19 -05:00
ameerj
5286a7bc4c shader_recompiler: Rename backend emit_context files 2021-12-05 16:33:44 -05:00
ameerj
41aec2773f general: Add missing copyright notices 2021-12-05 16:18:53 -05:00
Feng Chen
4dd85f86a8 Implement convert legacy to generic 2021-11-19 22:53:58 +08:00
Fernando Sahmkow
6f98690963 ShaderCache: Better fix for Shuffling gl_FragCoord 2021-11-16 22:11:33 +01:00
FernandoS27
de1c8c5c2c Texture Cahe/Shader decompiler: Resize PointSize on rescaling, refactor and make reaper more agressive on 4Gb GPUs. 2021-11-16 22:11:33 +01:00
ameerj
618de4e787 vulkan: Fix rescaling push constant usage 2021-11-16 22:11:31 +01:00
ameerj
581ea90062 rescaling_pass: Fix IR errors when unscalable texture types are encountered 2021-11-16 22:11:30 +01:00
ameerj
99eec162da rescaling_pass: Logic simplification and minor style cleanup 2021-11-16 22:11:30 +01:00
ameerj
276565973f rescaling_pass: Scale ImageFetch offset if it exists
Plus some code deduplication
2021-11-16 22:11:30 +01:00
ameerj
dd66384451 rescaling_pass: Enable PatchImageQueryDimensions on fragment stages 2021-11-16 22:11:30 +01:00
ameerj
b027fac794 gl_texture_cache/rescaling_pass: minor cleanup 2021-11-16 22:11:30 +01:00
ameerj
edb5844240 rescaling_pass: Fix and simplify shuffle/fragcoord pass 2021-11-16 22:11:30 +01:00
Fernando Sahmkow
b3a9c8f108 Shader: Don't rescale FragCoord if used by Shuffle 2021-11-16 22:11:30 +01:00
ameerj
68e038404c shader, video_core: Fix GCC build errors 2021-11-16 22:11:29 +01:00
ameerj
65781f88f8 emit_spirv: Fix RescalingLayout alignment 2021-11-16 22:11:29 +01:00
Fernando Sahmkow
dfa8291526 RescalingPass: Agregate pixels on texelFetch while on Fragment Shader 2021-11-16 22:11:29 +01:00
Fernando Sahmkow
8f78444de3 shader: Fix TextureSize check on rescaling. 2021-11-16 22:11:29 +01:00
ameerj
dc28284437 emit_spirv: Fix RescalingLayout alignment 2021-11-16 22:11:29 +01:00
ReinUsesLisp
e66d5b88a6 shader: Properly scale image reads and add GL SPIR-V support
Thanks for everything!
2021-11-16 22:11:29 +01:00
ReinUsesLisp
fc9bb3c3fe shader: Properly blacklist and scale image loads 2021-11-16 22:11:29 +01:00
ReinUsesLisp
cfeb161c7e glsl/glasm: Pass and use scaling parameters in shaders 2021-11-16 22:11:29 +01:00
ameerj
f086c82e1f gl_graphics_pipeline: Add downscale factor to shader uniforms 2021-11-16 22:11:28 +01:00
ReinUsesLisp
656adee630 spirv: Implement rescaling patching 2021-11-16 22:11:28 +01:00
ReinUsesLisp
01379c5e3c shader/rescaling_pass: Patch more instructions 2021-11-16 22:11:28 +01:00
ReinUsesLisp
c15332c44f shader: Add IsTextureScaled opcode 2021-11-16 22:11:28 +01:00
ReinUsesLisp
c892359d1b shader: Add copy constructor to instructions 2021-11-16 22:11:28 +01:00
ReinUsesLisp
95761cc6a7 shader: Add integer division opcodes 2021-11-16 22:11:28 +01:00
ReinUsesLisp
e580299467 shader: Fix rescaling pass 2021-11-16 22:11:28 +01:00
ReinUsesLisp
1672e9ba09 shader: Fix resolution scaling pass 2021-11-16 22:11:27 +01:00
ReinUsesLisp
fb924ea85c shader: Add resolution down factor opcode 2021-11-16 22:11:27 +01:00
Fernando Sahmkow
360e897ccd ShaderDecompiler: Add initial support for rescaling. 2021-11-16 22:11:27 +01:00
bunnei
30442d8a89
Merge pull request #7260 from vonchenplus/spirv_support_legacy_attribute_v2
shader: Spirv support legacy attribute v2
2021-11-14 02:07:45 -08:00
Feng Chen
f2a4204245 Simply legacy attribute implement 2021-11-04 09:26:16 +08:00
bunnei
7ce29ea74e
Merge pull request #7262 from FernandoS27/Buffalo-buffalo-Buffalo-buffalo-buffalo
ShaderCache: Order Phi Arguments from farthest away to nearest.
2021-11-02 16:34:03 -07:00
Fernando Sahmkow
2d4bbd83e6 Shader Cahe: Fix Phi Nodes on GLASM. 2021-11-02 00:59:25 +01:00
Fernando Sahmkow
194579bc4f ShaderCache: Fix Phi Nodes Type on OGL. 2021-11-01 22:26:17 +01:00
Fernando Sahmkow
c50ad56bf5 ShaderCache: Order Phi Arguments from farthest away to nearest. 2021-10-31 19:34:15 +01:00
vonchenplus
49f9a44235 Support gl_FogFragCoord attribute 2021-10-31 23:34:17 +08:00
Fernando S
7aa0d97eed
Merge pull request #7201 from ameerj/spirv-depth-sampling
emit_spirv_image: Fix depth image implicit lod sample in non-fragment stages
2021-10-30 16:45:58 +02:00
vonchenplus
36c21ff6cb Support gl_BackSecondaryColor attribute 2021-10-26 23:14:40 +08:00
vonchenplus
92bebecf46 Support gl_FrontSecondaryColor attribute 2021-10-26 23:09:44 +08:00
vonchenplus
b04c7b6343 Support gl_BackColor attribute 2021-10-26 23:01:57 +08:00
Fernando Sahmkow
e5291e2031 TexturePass: Fix clamping of images as this allowed negative indices. 2021-10-24 20:46:36 +02:00
ameerj
06894b0711 emit_spirv_image: Fix depth image implicit lod sample in compute
Ensures all drivers behave the same way in this case.
2021-10-17 17:09:11 -04:00
Ameer J
3791c7ca82
Merge pull request #7077 from FernandoS27/face-down
A series of fixes to queries and indexed samplers.
2021-10-16 21:41:54 -04:00
Fernando Sahmkow
3f4444b552 Shader Compiler: avoid overflowed indices on indixed samplers. 2021-10-17 03:38:09 +02:00
Morph
e29f3b87f1 style: Remove extra space preceding the :: operator 2021-09-29 01:26:01 -04:00
ameerj
73666fb262 general: Update style to clang-format-12 2021-09-24 15:52:05 -04:00
Fernando Sahmkow
8984abfc76 Spir-V: Rescale the frag depth to 0,1 mode when -1,1 mode is used in Vulkan. 2021-09-15 21:30:33 +02:00
Morph
9248442bb2
Merge pull request #6948 from ameerj/amd-warp-fix
shaders: Fix warp instructions on 64-thread warp devices
2021-09-12 13:53:29 -04:00
bunnei
7e9163779d
Merge pull request #6962 from vonchenplus/spirv_support_legacy_attribute
renderer_vulkan: Spirv support glsl  legacy attribute
2021-09-08 14:04:44 -07:00
Feng Chen
b1e655f898 Detail adjustment 2021-09-08 10:30:00 +08:00
Feng Chen
bbc1800c1b Detail adjustment 2021-09-08 09:53:10 +08:00
Feng Chen
e5ca733722 Re-implement get unused location 2021-09-07 13:22:52 +08:00
Feng Chen
9cdf2383e9 Move attribute related definitions to spirv anonymous namespace 2021-09-07 12:34:35 +08:00
Feng Chen
1de9e4e121 Dynamic get unused location 2021-09-06 10:46:03 +08:00
Feng Chen
d994466a08 Implement intput and output fixed fnc textures 2021-09-06 10:36:45 +08:00
Feng Chen
a7bbaa4897 Rename parameters 2021-09-03 23:52:20 +08:00
Feng Chen
cf26f375ff Fix create GraphicsPipelines crash 2021-09-03 22:55:53 +08:00
Feng Chen
1e2a89d306 Add input/output location 2021-09-02 23:34:51 +08:00
bunnei
b2572a56d3
Merge pull request #6900 from ameerj/attr-reorder
structured_control_flow: Add DemoteCombinationPass
2021-09-01 17:36:26 -07:00
ameerj
d956fb3c7c emit_glsl_warp: Fix shuffle ops for 64-thread warp sizes 2021-08-31 16:11:25 -04:00
ameerj
5b45dfe971 emit_glsl_warp: Fix ballot related ops for 64-thread warp sizes 2021-08-31 16:11:25 -04:00
ameerj
a5d9dcf3d9 emit_spirv_warp: Fix shuffle ops for 64-thread warp sizes 2021-08-31 13:40:39 -04:00
ameerj
95213270ef emit_spirv_warp: Fix ballot related ops for 64-thread warp sizes 2021-08-31 13:40:12 -04:00
Feng Chen
73b11f390e Add colorfront and txtcoord support 2021-09-01 00:07:25 +08:00
ameerj
907dfbea71 structured_control_flow: Skip reordering nested demote branches.
Nested demote branches add complexity with combining the condition if it has not been initialized yet. Skip them for the time being.
2021-08-30 11:46:25 -04:00
ameerj
4fda7f1c82 structured_control_flow: Conditionally invoke demote reorder pass
This is only needed on select drivers when a fragment shader discards/demotes.
2021-08-30 11:46:24 -04:00
ameerj
862dc2b2b3 structured_control_flow: Add DemoteCombinationPass
Some drivers misread data when demotes are interleaved in the program. This moves demote branches to be checked at the end of the program.
Fixes "wireframe" issue in Pokemon SwSh on some drivers
2021-08-28 11:35:25 -04:00
ameerj
6e407c02d8 emit_spirv_context_get_set: Fix Get FrontFace return value
The IR expects GetAttribute to return an F32 value. This case was returning a U32 instead.
2021-08-26 21:37:34 -04:00
Valeri
beb7305b73
SPIR-V: Merge two ifs in EmitGetAttribute 2021-08-19 20:13:46 +03:00
Morph
db07ca6c7f
Merge pull request #6767 from ReinUsesLisp/fold-float-pack
shader: Fold UnpackFloat2x16 and PackFloat2x16
2021-07-30 02:07:52 -04:00
bunnei
a98f14e9b0
Merge pull request #6722 from ReinUsesLisp/xmad-opts
shader: Fold integer FMA from Nvidia's pattern
2021-07-29 18:45:37 -07:00
ReinUsesLisp
8c9febe8f7 shader: Fold UnpackFloat2x16 and PackFloat2x16
Simplifies the code a bit when possible. These instructions should be
no-ops codegen wise.
2021-07-29 21:22:52 -03:00
ReinUsesLisp
1bb46b7d64 shader: Mark ConvertF16F32 and ConvertF32F16 as fp16 instructions
Fixes instances where fp16 types are not declared on SPIR-V but they are
used. This shouldn't happen on master, as it's been uncovered by an
additional optimization pass.
2021-07-27 21:33:05 -03:00
Lioncash
c27ddb44de exception: Make constructors explicit
Ensures that exception construction is always explicit.
2021-07-27 04:15:14 -04:00
Lioncash
e490ddf327 exception: Make what() member function nodiscard 2021-07-27 04:14:32 -04:00
Lioncash
90f3678ada exception: Narrow down specific header
We can use the <exception> header instead of pulling in all of the
exception-style classes.
2021-07-27 04:09:18 -04:00
Rodrigo Locatti
c0f99558fb
Merge pull request #6724 from lioncash/nodisc-shader
shader_recompiler: Remove unnecessary [[nodiscard]] instances
2021-07-26 16:35:21 -03:00
Rodrigo Locatti
de0b89792c
Merge pull request #6726 from lioncash/hguard
emit_spirv_instructions: Add missing header guard
2021-07-26 16:35:11 -03:00
Rodrigo Locatti
3d97f1e6cf
Merge pull request #6727 from lioncash/topology
emit_glasm: Fix LINESS_ADJACENCY typo in InputPrimitive()
2021-07-26 16:35:03 -03:00
Rodrigo Locatti
b2b3fcdccd
Merge pull request #6723 from lioncash/shader
object_pool: Add missing return in Chunk move assignment operator
2021-07-26 06:01:21 -03:00
Lioncash
3e7813e49d emit_glasm: Fix LINESS_ADJACENCY typo in InputPrimitive()
This should be LINES_ADJACENCY
2021-07-26 04:44:56 -04:00
Lioncash
c2915d9f2f emit_spirv_instructions: Add missing header guard 2021-07-26 04:28:35 -04:00
Lioncash
06ca911621 shader_recompiler: Remove unnecessary [[nodiscard]] instances
[[nodiscard]] doesn't do anything on functions with a void return type
and causes superfluous warnings.
2021-07-26 04:23:59 -04:00
Lioncash
0b67df1f7c control_flow: Fix duplicate switch case in OpcodeToken
This previously duplicated the case of the PBK case above it.
2021-07-26 04:16:34 -04:00
Lioncash
89ad9df0e9 object_pool: Add missing return in Chunk move assignment operator
Prevents undefined behavior from occurring.
2021-07-26 04:01:05 -04:00
ReinUsesLisp
66a0cedba3 shader: Fold integer FMA from Nvidia's pattern
Fold shaders doing "a * b + c" on integers from the pattern generated by
Nvidia's GL compiler.

On a somewhat complex compute shader it reduces the code size by 16
instructions from 2 matches on Turing GPUs.

On Intel as extracted from KHR_pipeline_executable_properties:
Before the optimization:
```
Instruction Count: 2057
Basic Block Count: 45
Scratch Memory Size: 14752
Spill Count: 232
Fill Count: 261
SEND Count: 610
Cycle Count: 11325
```

After the optimization:
```
Instruction Count: 2046
Basic Block Count: 44
Scratch Memory Size: 13728
Spill Count: 219
Fill Count: 268
SEND Count: 604
Cycle Count: 11367
```
2021-07-26 04:58:02 -03:00
ReinUsesLisp
09fb41dc63 shader: Use TryInstRecursive on XMAD multiply folding
Simplify a bit the logic.
2021-07-26 04:15:27 -03:00
ReinUsesLisp
f6f0383b49 shader: Add TryInstRecursive utility to values 2021-07-26 01:31:05 -03:00
ReinUsesLisp
7f13104c17 shader: Support out of bound local memory reads and immediate writes
Support ignoring immediate out of bound writes. Writing dynamically out
of bounds is not yet supported (e.g. R0+0x4).

Reading out of bounds yields zero. This is supported checking for the
size from the IR; if the input is immediate, the optimization passes
will drop it.
2021-07-22 21:51:41 -04:00
ameerj
56478bc9ac shader: Fix disabled attribute default values 2021-07-22 21:51:40 -04:00
ameerj
56c30dd9e0 glsl: Simplify FCMP emission 2021-07-22 21:51:40 -04:00
ameerj
79d2684261 glsl: Update TessellationControl gl_in
Adheres to GL_ARB_separate_shader_objects requirements
2021-07-22 21:51:40 -04:00
ameerj
fc7bed21b5 shader: Implement ISETP.X 2021-07-22 21:51:40 -04:00
ReinUsesLisp
bf2956d77a shader: Avoid usage of C++20 ranges to build in clang 2021-07-22 21:51:40 -04:00
ameerj
94af0a00f6 glsl: Clamp shared mem size to GL_MAX_COMPUTE_SHARED_MEMORY_SIZE 2021-07-22 21:51:40 -04:00
lat9nq
49946cf780 shader_recompiler, video_core: Resolve clang errors
Silences the following warnings-turned-errors:
-Wsign-conversion
-Wunused-private-field
-Wbraced-scalar-init
-Wunused-variable

And some other errors
2021-07-22 21:51:40 -04:00
ReinUsesLisp
2235a51b5d shader: Manually convert from array<u32> to bitset instead of using bit_cast 2021-07-22 21:51:40 -04:00